Aider
AI pair programming in your terminal, with git-native workflows.
About
Aider is an open-source terminal pair-programming tool that edits code across a repository with an LLM, using repo maps and structured edit formats (search/replace, diff, whole-file).
It works with cloud providers (Anthropic, OpenAI, DeepSeek, and many more) and local models, integrates with git for automatic commits, and runs configured linting and test suites after changes.
It installs via pip/aider-install or a one-line shell script, is Apache-2.0 licensed, and had roughly 49.5k GitHub stars as of October 2026.
What it can do
- Terminal-based pair programming that edits across whole repositories
- Repo map built from tree-sitter AST parsing for codebase-wide context
- Structured edit formats (search/replace, unified diff, whole-file) tuned per model
- Supports 100+ programming languages
- Works with cloud LLMs and local models; Architect/Editor dual-model mode
- Git integration: automatic commits with sensible messages, diff/undo via familiar git tools
- Automatically lints and tests after changes and fixes detected problems
- Accepts images and web pages as context; voice-to-code input
- Works alongside any editor via comment-based requests
Setup
- Install aider in an isolated environment: run
python -m pip install aider-installthenaider-install(works with Python 3.8-3.13; it installs its own Python 3.12 environment if needed). One-liner alternative on macOS/Linux:curl -LsSf https://aider.chat/install.sh | sh; on Windows:powershell -ExecutionPolicy ByPass -c "irm https://aider.chat/install.ps1 | iex". - Set an LLM API key in your environment (e.g.
export ANTHROPIC_API_KEY=...) or pass it on the command line with--api-key anthropic=<key>. - Change into your git repository:
cd /to/your/project. - Launch aider with a model, e.g.
aider --model sonnet --api-key anthropic=<key>(other examples from the docs:--model deepseek,--model o3-mini). - Chat in the terminal to request edits; aider applies them, commits via git with sensible messages, and runs your linters/tests automatically.
Prerequisites
Python 3.8–3.13; an LLM provider API key (or a local model endpoint)
Use cases
Works with
Trust score — 8/10
Very active (commits adding new model variants in Oct 2026), ~49.5k stars, Apache-2.0, long-lived since 2023 with no dead-project signal. Caveats: opt-in anonymous analytics via PostHog exists (independently audited: no code/prompts/keys collected, `--analytics-disable` opt-out works); the README's "works best with" model list names older models (Claude 3.7 Sonnet) while commits track 2026 models — docs lag the code slightly. Metrics-v1 score 8/10 — adoption 8.5, maintenance 4.4, openness 10, manual 9.
★ 49,502 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.5/10
- Maintenance · recency, releases, issue load
- 4.4/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · 49,502 GitHub stars (log-scaled vs catalog peers)
- · blended with ~260,509 monthly package downloads
- · last push 143d ago; 0 releases/12mo; 1913 open issues
- · manual review: opt-in analytics independently audited; README model list lags commits
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.